Transit Signal Priority: Help or Hype? Peter G. Furth Northeastern University

Transcription

1 Transit Signal Priority: Help or Hype? Peter G. Furth Northeastern University

2 Transit Signal Priority Help or Hype? Zurich: nearly zero traffic delay and impeccable punctuality for trams Portland, San Diego: green wave for light rail station-to-station Yet, many US applications: < 2 s savings per intersection, or No measurement at all 2

3 Transit Priority as a Societal Objective Transit use benefits society in terms of Congestion Air quality, climate impact, energy use Vibrant communities Priority breaks the vicious cycle in which congestion drives people to switch from transit to car 3

4 Priority Makes Sense One extreme ($$$$): build a metro Other extreme: do nothing, buses become swamped in congestion Traffic delay can represent 30% of a bus route s operating cost Feeds vicious cycle of ever-lower transit use In between: Priority in space: bus lanes, etc. Priority in time: signal priority 4

5 What Priority Means to a Transit Operation Reduced running time ($$, passenger benefit). Measures: Delay per intersection Average route running time Improved reliability ($$, passenger benefit). Recovery time, cycle time based on 95-percentile running time Tri-Met Line 12: reduced needed cycle length from 104 min to 93 min (11%) 5

6 Operational Control: Schedule Adherence, Crowding Schedule deviation along the route, without priority, Eindhoven With priority 6

7 Priority Makes Transit More competitive Bus has natural disadvantages due to stops and walking / waiting Priority compensates, especially if cars suffer congestion More socially acceptable (red carpet) my great aunt 7

8 Levels of Priority in Modern Traffic Signal Controllers Highest level (preemption) reserved for emergency vehicles and railroad crossings Pre-programmed priority tactics include green extension and early green More intelligent tactics require custom programming, usually with central control & communication 8

9 cumulative vehicles Green Extension Built-in logic in modern controllers Large benefit to a few buses Therefore little disruption to traffic Extension increment is a parameter (10 s? 15 s?) Without priority cumulative vehicles With green extension bus delay bus delay v 1 s 1 v 1 s 1 red green time X normal red green 9 time

10 cumulative vehicles Priority Push is not the same as Without priority Allowed Extension cumulative vehicles With green extension bus delay bus delay v 1 s 1 v 1 s 1 red green time X normal red green time E[ delay] = 2 R 2C 1 1 v s 2 2 = R 1 R X E[ delay] X (1 v s 2C 1 v s ) C 2C Priority push!

11 Priority Push vs. Green Extension (cycle length = 100 s, red time = 50 s, degree of saturation = 85%) Push (s) Green Extension (s)

12 Priority Push vs. Red Time for allowed green extension of 15 s (cycle length = 100 s, degree of saturation = 85%) Push (s) Red Time (s)

13 Detection Sketch 1 Check-in detector location Early enough to allow time to respond Late enough to estimate bus arrival time Checkout detector to cancel request Avoid wasted green Performance measurement Fixed detector (buried antenna) AVL plus communication with the signal control box can serve as a detector 13

14 Upstream Detector for Green Extension: travel time = maximum green extension Simplicity: Request = detection Weaknesses: assumes constant speed no flexibility for updates, time of day settings may not be suitable for other priority tactics 14

15 What if There s a Near-Side Stop? Detector located just after stop Disable optical signal until door closes (Portland, OR) Possible need for priority against a vehicle queue blocking entry to the bus stop 15

16 Common Weaknesses in Signal Priority Implementation 1. Cautious priority: Inhibit priority for 5 minutes or 1 full cycle after priority interruption if cross street occupancy exceeds threshold 2. Unnecessarily hurting traffic (leading to cautious priority) Lack of checkout detector = wasted green Lack of compensation creates large queuing impacts Inflexibility in standard coordination logic 3. Apply easy, rather than intelligent, tactics 4. Lack of data collection and analysis Nobody ever gets it right the first time Taking the objective seriously 16

17 Fully actuated (uncoordinated) signal control naturally compensates Fully actuated = hold green until the queue clears (subject to min, max green) Interrupted phases get longer green in the next cycle, usually enough to clear the queue Transit phase Priority green allows extra cars to be served Reduces demand for green time in the next cycle System quickly recovers 17

18 Lack of compensation in coordinated systems Fixed cycle length Fixed point = end of coordinated phase Uncoordinated phases may run shorter than their allotted split, but not longer Coordinated phase gets the slack time (starts early) SBT SBL NBL 4 EBT SBT EBL EBT WBT WBL EBL WBL WBT NBT NBL NBT SBL 8 E-W Street Suppose coordinated phase is 2 (EBT) Green extension for 2 no mechanism for compensation Green extension for 4 slight compensation N-S Street 18

19 Returning to Coordination In coordinated systems, recovery means dissipating queues AND Returning to the background cycle Short way = shorten phases following an extension Long way = lengthen phases, skip a cycle 19

20 Early Green Truncate and possibly skip preceding phases What s the truncation rule? How much to shorten competing phases? Can they be skipped? Smaller benefit to large number of buses More traffic impact; hard to implement when bus frequency is high 20

21 Early Green Issues Exclusive lane for bus? (No queue, easier arrival time prediction) Mixed traffic: Eindhoven s electronic bulldozer Arrival time prediction How far upstream? (Bus stop and intersection spacing ) Tracking queue length to know how long is needed to flush out the queue (Zurich s trap logic) 21

22 Early green (truncating competing phases) under coordination NBL 4 EBT WBL SBT EBL 5 WBT NBT 6 7 SBL 8 E-W Street Assume phase 2 is coordinated N-S Street Early green for 2 is possible, but without compensation to shortened phases Early green for 3 not possible under standard logic 22

23 Eindhoven Experiment Coordinated phase (ring road, no buses) Buses every 10 min NB and SB 23

24 Existing Priority Scheme Coordinated phase is 6 Buses are on 4 and 8 (every 10 minutes) Priority only if bus is more than 20 s late (about half the buses) NBL 4 EBT SBT WBL EBL WBT NBT SBL 8 E-W Street N-S Street Green extension Aggressive early green: reduce intervening phases to minimum green Short way minimum green to return to background cycle 24

25 Intersection Experiment and Site Description One day each of priority to all buses (absolute) priority to late buses (conditional) no priority Entry Exit Exit Entry Entry Camera facing each approach Film trap between entry & exit Playback in lab Exit Exit Entry

26 Experiment Eindhoven Average Transit Delay [sec] Delay Public Transport North Bound South Bound North Bound South Bound No Priority Abs Priority Cond priority AM PM

27 Experiment Eindhoven: Traffic Impact Average Vehicle Delay [sec] Delay Time No Priority Abs Priority Cond Prioriy AM PM

28 Experiment Eindhoven Average Vehicle Delay per Approach [sec] Delay Time No Priority Abs Priority Cond Priority 20 0 West Bound North Bound Eas t Bound South Bound West Bound North Bound Eas t Bound South Bound AM PM

29 Experiment Eindhoven Relative Capacity per Approach (no priority = 100%) No Priority Abs Priority ond Priority Eastbound Westbound Northbound Southbound Northbound Southbound Eastbound Westbound AM PM

30 Lessons from Eindhoven Aggressive early green resulted in near-zero delay for buses Conditional priority needs finely-tuned schedule Schedule too tight bus always late absolute priority Schedule too loose bus always early no priority Lack of compensation: OK for 6 interruptions per hour, but not 12 Capacity loss due to Early green truncations, but more from short way recovery to background cycle 30

31 Early Red Shorten bus street s current green to get faster return to green in the next cycle Needs advanced detection (almost a full cycle) Incompatible with typical coordination logic custom programming 31

32 Green extension to clear queue from bus stop Force signal to red during stop Minimizes bus s impact on road capacity Return to green as quickly as possible Flush-and-Return Early green tactic for Near-Side Stops Tested using simulation on San Juan (PR) arterial 32

33 Phase Rotation Example: change leading bus phase to lagging Lagging bus phase becomes leading like early green Leading bus phase becomes lagging like green extension (more effective) Used extensively in Germany Zurich s pre-application safety campaign: random phase sequencing for 6 months! 33

34 Phase Insertion Second realization on bus detection only Greatly reduces red period for bus big reduction in delay Zurich s insert and return 34

35 Passive Priority Treatments that favor buses but don t rely on bus detections Favorable splits for bus phase Favorable offsets (progression) for bus Hard to do over more than a few intersections due to uncertain dwell time Short cycles or double realizations (short red is the key) 35

36 Ruggles Bus Terminal Study with Burak Cesme, PhD student at Northeastern Univ. Using VISSIM simulation and VAP signal control programming 36

37 Main Entrance 1 Ruggles- Busway Exit 4 Back Entrance Cass-Columbus 2 Ruggles- Tremont-Whittier 3 Tremont-Cass 37

38 38

39 39

40 Bus Delays with Incremental Priority Treatments, by Route 40

41 Passive priority: more time to bus left turn. Note: 5 s increase in split consumes only 2.5 s! Proportion Max Green = 16 seconds Green Time ( s ) Proportion Max Green = 21 seconds Green Time ( s ) Avg Green (EBL) = 15.3s p (max-out) = 84.6% Avg bus delay = 98 s Avg Green (EBL) = 17.8s p (max-out) = 51.2% Avg bus delay = 67 s 41 41

42 Intelligent Green Extension: 10 s extension for cost of 0.5 s Proportion No Priority Green Time ( s ) Proportion With Green Extension Extended Green Green Time ( s ) Avg Green (EBL) = 17.8s p (max-out) = Avg Green (WBT) = 30.3s Avg Green (EBL) = 18.1s p (max-out) = p (extended) = Avg Green (WBT) = 29.8s 42

43 Green Extension Only 0.30 Inserting 10 s phase: only consumes 2.5 s Primary Phase, when Phase Insertion is Programmed 0.30 Proportion Green Time ( s ) Extended Green Proportion Green Time ( s ) Extended Green Avg Green (EBL) = 18.1s Bus delay = 55 s Avg Green (WBT) = 29.8s p (insertion) = Avg Green (primary) = 14.7s Avg Green (insertion) = 4.4s Avg Green (total) = 19.1s Bus delay = 33 s Avg Green (WBT) = 27.3s 43

44 Conditional Priority Priority to Late Buses Less interference with traffic (Eindhoven) Push-pull means of operational control (Einhoven) What is Late: 15 s or 3 minutes? Demands fine-tuned schedule Headway-based priority for short-headway service 44

45 Priority Queue Management Detectors & logic for queue management Stopped cars, not moving cars, hinder buses (Zurich) (Eindhoven) 45

46 Predictive Priority Remote, upstream detection: simulated on Huntington Ave, used in Salt Lake City Detector 1 used to predict bus arrival at 4 (~2 minutes advance) Adjust cycle lengths so that bus will arrive on green Last-minute priority as backup Adaptive (learning) algorithm for predicting bus arrival 46

47 Dynamic Coordination (Zurich) Small zones (1-3 intersections) No fixed clock Shape green waves through the zone around bus Zone boundaries are segments that offer storage buffer 47

48 Self-Organizing Coordination Each signal s start of green becomes a request to downstream signal Peer-to-peer communication between signals upstream signal s request has lower priority that bus request Result: spontaneous green wave Inherently interruptible 48

49 Multi-Level Priority (South Tangent = Haarlem Airport Amsterdam South) Bus is early: green extension only 0 to3 min late: normal early green More than 3 min late: aggressive early green (skip competing phases) 49

50 Six Keys to Performance 1. Aim for near-zero delay 2. Multiple intelligent and aggressive tactics, with compensation 3. Coordinate with scheduling (cond l priority) 4. Alternatives to rigid coordination 5. Advanced prediction with gradual cycle adjustments 6. Custom programming, performance measurement, & continual improvement 50